Filter section
The goal of the filter is to change the shape of the pulses. If no modification is done on the pulse shape, we would only be able to say when the pulse occurred with a precision not higher than the sampling period. Indeed we would be able to see a sample having one logic value and the following one with the other logic value. But it would be impossible to determine when did this change occur within the sampling period. Interpolation is required. To do this, the idea is to convert the rising/falling edge of the pulse in a longer observable event. In our case, we transformed the pulse falling edge into a damped sine wave, impulse reponse of an LC filter. Thanks to fitting techniques, it is possible to determine the phase of this signal and, consequently, its time of arrival. When the time of arrival of the different pulses are known, the delay between them can be computed by computing their time of arrival difference.
This page is dedicated to the built filter which converts the pulses edges into damped sines. The filter for one channel is represented in figure 1. The used operational amplifier is a LMH6624 and the Schottky diodes are HSMS-281C-BLKG. This filter is designed to convert the falling edges into damped sine. The diodes can be reversed to convert the rising edges. The schematics does not represent the decoupling capacitors around the operational amplifier. We used a combination of 1uF 100nF and 1nF X7R capacitors on both the positive and negative inputs to do the job.
The capacitor C1 is used to realise a high-pass with the rest of the circuit. On the rising edge, the capacitor C1 is charged through the D1 and R2. D2 is in off-state and nothing arrives to the filter output.
In contrast, on the falling-edge, the pulse discharges C1 through the D1, R3 and the filter. The shape of the impulse arriving in the LC-filter depends on the voltage difference of the pulse, the fall time and the size of the ratio C1 to C2. The R3 value is 100 Ohm to have an equivalent 40 Ohm input impedance on each pulses and then avoid reflections.
The operational amplifier is used as both a buffer and an amplifier. Its gain is set to 2. The resistor R4 is added for the amplifier stability. The damping factor of the filter does not seem to be affected by the addition of this resistor.
The actual filter is built with a nickel-zinc ferrite coil with a 1 uH inductance and a 100 pF capacitor X7R.
It is also mandatory to be independant of the incomming signal shape and level. To do that, we use CMOS gates (5PB1108). On one hand they will always give the same output once the input becomes higher than the threshold level, we will then always have the same damped sine shape. On the other hand, they clean the edge. As the LC impulse response is linked to the quality of the pulse, we clean then the damped sine shapes and improve the fitting results.
Despite the CMOS gates, we still have distortion at the beginning of the pulse (Figure 2). This ramp is due to the fact that our pulse is not a perfect Dirac model. This ramp adds error on the measurements and has to be removed before applying the fitting algorithm. The proposed solution consists in removing these points from the analysed frame before applying the fitting. By doing this, we unfortunately loose some SNR (Signal to Noise Ratio) points. We tried to increase the amplifier gain to have increase the SNR of the points in the used frame. Unfortunatelly, this saturates the ADC. As its recovery time is relatively long, the non linearities decrease the precision.
Figure 2 - Distorrtion at the beginning of the pulse. Resolution is
20ns/DIV on the x axis and 227 mV/DIV on the y axis
Other filers
Figure 2 presents two other filters that could be used instead of the actual one.
Figure 2 - Other possible filters
18th of January 2016 - Nicolas Boucquey